Machine learning algorithms have been behind the scenes of modeling and ad placement for years but now tools like ChatGPT, MidJourney, DallE, and more have put AI in our hands.
Learn what nonprofits are doing with this new technology, the promise and the perils, and what prompts and tools you can use to make your life easier. Slides are being developed for this event.
Presented By The NonProfit Times and the Resource Alliance
Panel:
• Nick Ellinger is chief brand officer at Moore. He joined Moore in January 2020 where he works to create great experiences for nonprofits and their constituents.
• Filipe Páscoa is a data crafter and senior fundraising and public engagement strategist. One of the pioneers in using artificial intelligence (started using GPT in 2022) and social data intelligence in the sector.
Moderator:
• Paul Clolery is the Editor-in-Chief of The NonProfit Times.
Stream this Webinar here!
View Slides here!
Transcript:
0:04
Good afternoon, and welcome to Generative AI & Your NPO: How ChatGPT+ Will Change The Way NPOs Work into the Future. I’m Paul Clolery, Vice President and Editorial Director of the non-profit Times.
0:22
And Hi.
0:24
Oh, go ahead.
0:26
Yeah, hi everyone, my name is …, I’m the CEO of the Resource Alliance, and actually dialing in from the other side on some based in Scotland that moment. So, really nice to do this together with the non-profit Times.
0:37
A little bit about the resource Lines, in case you don’t know us, we form a Global Community of over 50,000 …, campaign ish Changemakers, to make the world a better place and we try to do a lot around capacity building, sharing of knowledge, new innovative thinking, and this webinar is, of course, the whole example of it. Some of you might know us from our flagship event which is the IFC, the International Congress that is held every year in October, in The Netherlands, but we really wanted to share more of that knowledge and expertise throughout a year.
1:10
I’m very happy to be present here with all and the non-profit times in this webinar and welcoming all of you. Of course, will bring a combined audience, is here, that means audience from the US, but also from a course, our global community. So, please share in the chat where you’re dialing in, who you are always nice to know.
1:33
Yeah, well, yeah, In the room today, and who are listening in this session.
1:37
And, of course, keep it informal, so the special questions are happy to answer them, if possible, Acting.
1:47
We’re really excited about this collaboration between the non-profit Times and the Resource Alliance.
1:53
And we’ve got some other webinars cooking for the rest of the Year two. So, this is really, really exciting for us, So, let’s just get started here. We’ve got our two terrific panelists, Nick Ehlinger, as Chief Brand Officer, and more Nick, Give them a wave.
2:09
Join more in January 2020 where he works to create great experiences for non-profits, other constituents, but they’ve been around this block quite a bit. He’s been in the industry a very long time. And he’s very knowledgeable about what’s going on in, in, not just in the US, but in tech and everything that has to do with fundraising. And Felipe: Pascal is the data crafter, and a senior at fundraising.
2:35
Senior Fundraising, Public Engagement Strategist, is one of the pioneers in using artificial intelligence, started using …
2:43
in 20 22, and as a social data intelligence, and social data intelligence in the sector. So, we’re thrilled to have both of these gentlemen here. And we’ll be checking in with your questions, and our questions, as we go along. And hope that everyone on the panel will be mixing it up, as well.
3:00
So, I want you to give, give, give it a start.
3:06
Thank you, so much, for joining today.
3:08
I wanted to start by doing a quick definition of terms, since this is a relatively new area for a lot of folks, there’s We’re going to be talking about artificial intelligence. We’re going to be talking about machine learning. And we’re going to be talking about action over here.
3:25
So artificial intelligences, anything that really aims at creating a human mimicking, human intelligence.
3:32
Um, within that, with, within that, is machine learning. So, not all artificial intelligence as machine learning.
3:40
Machine learning is a subset where you have machines that are improving with experience.
3:46
So, for example, you can write a program that has a whole bunch of if then statements.
3:51
If they say this, you say that as a way of generating discussions. That’s artificial intelligence trying to mimic what a human would do.
3:59
It’s not machine learning because there’s not a learning component, so a machine learning system, pardon me.
4:07
Redundancy, a machine learning system learns.
4:09
Then, within machine learning, there’s something called generative AI. It’s called Generative, because we’re generating some content from a trading base.
4:18
And so, what you have when you put in generative AI, we’ll show an example of this, is where you have a set of training data come in do A neural network.
4:29
And then you prompt that with an input and say, here’s what I’d like to get based on this training data, and it will create the best version of that out.
4:38
So we’re talking about generative AI, and it’s important to note that some of the most useful machine learning systems right now isn’t generating.
4:48
It isn’t chat, GBT, addition, some of these systems, it’s taking all of the data that we have out there, all of the things that we could possibly know about a person identifying it with that person, and then activating on that to have extremely personalized, customized, and targeted communications.
5:07
So when I started in non-profit marketing, back in the early mesozoic, as Paul mentioned, you would use RSM analysis, recency, frequency, monetary value, That’s good transactional information, but we have so many data points on so many people right now that even in 2010, there are studies showing that a machine learning system, using the various possible data, could get you better results than strictly looking at transactions Numbers of study.
5:39
In Germany, with one of the cash, like diastase, That, using a machine learning system, and Google location data, elevation location near nearest church.
5:51
And those sorts of variables are more predictive than transactional value, and gave me 12% better results for when you’re mailing premiums versus when you’re not than any other system.
6:03
And so, that’s based on 13 year old technology.
6:07
We are already doing much better in our targeting, and we can do better in our personalization and customization, but the new thing that we’ve been talking about more and more, especially since the launch of G.t.p.
6:19
four in December is generative AIs.
6:24
And the basic idea is training data comes in one place.
6:28
That can be So for …
6:31
how do I get all of the information that I can from the internet That’s considered safe in to train about how people speak and also give us a knowledge base.
6:42
Then you have a prompt coming in. Here’s what I would like to hear about.
6:47
I would like I would like to hear about non-profit fundraising letter but write it in a sonnet form. those sorts of things and then it generates an output coming in.
6:58
That’s essentially what a change generative AI is, and if you are interested in any of these, there is a site called Bassanio. There’s an AI for that and when I took the screenshot, it’s probably more by now.
7:11
There are 5800 AIs that run 6800 different tasks, and it’s all sorts of different things that you have in there. You have ones that dude, test, like ….
7:24
There are ones that do image, like Mid journey or Gali, or those sorts of things.
7:29
There are videos. There’s code generation.
7:32
And so you can code with writing the things that you want to have happen, and it will translate it into Python, or SQL, or whatever language you need it to go into.
7:44
And it’s an extremely effective way of bootstrapping your way into her programming. There’s music now that’s coming in and people have generated additional songs by the Beatles.
7:57
Or do this in the style of trade or those sorts of things. All of, these have different sorts of regenerative APIs.
8:04
And the classic way of looking at this is through a Turing Test named for Alan Turing Decoder of the … machine during World War II and many other accomplishments. And he said you did this thought experiment that he called The Imitation Game.
8:19
Which is if you put a human, and I’m gonna Shean in two separate rooms so that an outside observer can’t.
8:28
And they’re typing responses back and forth. So you can’t tell from tone of voice or those sorts of things.
8:32
He was just a human, If that observer can’t tell the difference between the human and the artificial intelligence, can you retrieve, achieve true artificial intelligence?
8:45
And we’re seeing limited values of data right now.
8:50
G.t.p., for, which is the basis of what you’ve seen thing, what you’ve seen, …, when it’s the Open AI model that you see a lot of applications built up, It’s passed the bar exam.
9:02
And if it can learn to speak, like a lawyer can speaking, like a human, really be that far behind.
9:08
Also, here in the United States, we have advanced placement exams, which give you college credit, and it’s passed all, but one of the Advanced placement exams.
9:17
It gives you a 14, 10 on the sat, which is our college entrance exam here in the United States, and then going to what researchers look out for creativity.
9:28
It scores at the 99th percentile of how creative.
9:33
It is an idea generation. Now it does that in part by generating a lot of ideas. Some of those very rote. But some of those can be very creative as well.
9:44
So the basics for this, if you want to try it out for the first time and you haven’t, go to …, doctor open AI dot com.
9:53
You can do, or you can use Bing or those sorts of prompt engines, and you’re looking for a few things in your prompt.
9:59
You’re saying, Here’s how you should act.
10:03
Here’s the task I want you to do.
10:05
The constraints that I want you to put under the topic that I want to tone, the tone that I wanted for the audience that I’m writing for, and the context for it.
10:14
And so, an example of this would be act as an expert on marketing, and right and in 600 word blog post about …, in an informative tone, are non-profit marketers, who are looking to cut costs.
10:30
And that prompt will give you about a 600 Word blog post with those boxes checked off.
10:37
And so, and oftentimes that result will look very humid.
10:44
And there is a professor whose name is escaping me, I apologize.
10:50
Who goes in and he has what he calls the Wizard Post, you can, then, this is a chat engine. So you can go in and prompted for, more to say, OK, that’s a good idea.
11:00
But now, Pris.
11:01
Pretend like you’re a Wizard and you’re even better than the engine that you had before. Right, a better blog post under these parameters and it will give you literally a better post.
11:11
It will break through some of the things that were more rote to begin with.
11:16
And we’ll bring the knowledge up, and so you can take this is your initial prompt, but you can prompt and get it to give you more in the areas that you wanted to.
11:28
So, a few examples of this. one of my colleagues said, let’s take one of these image generating engines and say, What would happen when Nick looks Alker? So it generated this picture of me.
11:39
When I’m older, there was another one of these and if you put the side-by-side with my dad, it’s it’s pretty good.
11:45
So, you have these. And then, an example from the non-profit side of things, as we did some work for mountain furniture, the home of George Washington.
11:53
Who here in the States are first Precedented patriotic leader?
11:57
High treason, if you’re in the UK, I guess that would be, but, uh, here’s a video.
12:03
created Owly from AI, in terms of the script, and in terms of the images that go into it, with the human design, are helping to assemble. Greetings, Fellow Patriots and citizen of this great nation.
12:15
Your incredible dedication and proven leadership in supporting the Mount Vernon Ladies’ Association mission to preserve, restore, and manage my beloved home and property. Humbles me greatly.
12:28
Thanks to your incredible generosity, millions of people from all parts of the country and around the world, come to Mount Vernon to be informed and inspire. I can’t thank you enough.
12:38
I think, you’ll also be pleased to hear that Martha is absolutely delighted with all the work your partnership is making possible and, you know what they say, A happy wife means a happy afterlife.
12:49
As you know, this work is not government funded and we rely solely on the support of partners like you.
12:55
Remember, your gift is an investment in our youth and the future of our nation.
13:00
So, please be generous.
13:02
Thank you.
13:05
Hopefully, some of you are inspired by that and we’ll go to Mount vernon dot org to donate in the honor of our first precedent, and complete the Continue the restoration and the tours of that home.
13:21
So, what are the downsides, because it sounds like, oh, this isn’t magical tool, but there are definite downsides.
13:29
So Amazon tried to use a Machine Learning Engine to do its hiring, and said, Let’s go through all of the resumes that we have, see who we’ve hired, see who has worked out, and try to find more people like that.
13:48
The trick was, if you put in a resume, one from a man and one from a woman, it would choose to be his resume, OK. Bias in the system, what can we do about that?
14:00
Said machine learning system don’t take **** into account.
14:06
So then, the machine learning system went and they would still picked me all the resumes or female resumes. And it was what it was doing.
14:14
Is it was looking at whether you were in a fraternity or sorority. It was looking at whether you played men’s or women’s field hockey, for example.
14:23
It found a ways to discriminate because that was the training data that had had it, and said, OK, let’s take all of that out of the system route.
14:31
Jill found a way to discriminate between men and women, because men and women use different word choices and different verbs in their resumes.
14:42
And so, when you have biased training data coming in, because Amazon was doing what it had done, and because its hiring systems in the past had some bias to them.
14:56
Even any limited amount of bytes, the machine is trying to give you the best system to replicate that it’s fitting to the training data that it adds.
15:07
And so you have to monitor and fix for this type of bias and have all sorts of different training days.
15:17
We can imagine that in the non-profit sphere, let’s say that you had a file that was built entirely by premiums.
15:26
Well, it’s going to learn that premium pieces do better, It’s going to learn that, donors that respond to premiums are better donors, because those are the training data set, it hats.
15:36
And so it is going to be far more difficult to get a non premium donor matched up with a non premium piece. Even though that could be a great opportunity.
15:47
It, because the system will discount that, that doesn’t sound bad audience.
15:53
But mentally replace premium and non premium with white and black.
16:00
And you can see how this type of bias would change the way that your file works over time as you’re getting hundreds.
16:08
So that’s one challenge.
16:10
Another one is that these systems are trying to create, um, the approximation of human speech.
16:19
They are not trying to find out what’s true.
16:22
So, quick bit about me before I was more, I worked at Donor Voice for three years and that I was 14 years before that at mothers against drunk driving doing fundraising.
16:30
So I said, check GPG, who is Nicholas Ehlinger?
16:35
For one of the Go arounds. I was Vice President of Marketing Strategy at Bullying and Sonoma and I also worked at Marketing Democracy, a consulting firm. OK, let’s regenerate that, that’s clearly not true.
16:46
I’m an American political strategist, and I work for the VP of Classy.
16:51
And another one said, oh, he worked at change dot org.
16:55
Another one said I was at Precision dialog. Another one said that I was Chief Marketing Officer for every time for gun safety, and you can look at that.
17:03
And you can say, Oh, those are things that Nick Eleanor type person would do, but they’re not things that I did. And so, if you have truth, is of value, and, honestly, you should.
17:16
That’s a problem.
17:17
And then, brand safety is a huge problem.
17:20
Uh, we saw how people were placing ads next to, uh, anti vax or content, anti election, or election fraud base content AI is creating these content farms, and so, as you’re doing advertising and you’re placing advertising using machine learning, you also have to tell it, don’t put it on places that lie.
17:42
What do I want my brand associated with? So the things that we need to be doing to solve this is, we need to set goals that are very human goals that are above and beyond. Does this fundraise?
17:54
You have to have training data that encompasses all types of human experience. You want to be …, pursuing new audiences and communications.
18:02
if you have a problem with non premium communications for communications, too, a Hispanic audience, or whatever that is, you need to be the ones writing that non chachi PT and putting that in there.
18:13
And you have to have humans in the loop right now to solve for truth, and to make sure that you’re not having biased language.
18:22
Now, I’m going to kick it over.
18:27
Before we go to LA.
18:29
Thanks for everybody for joining in. We’ve got people from Chicago. We’ve got somebody from Cape Town, South Africa. We’ve got sunny Florida. We’ve got Massachusetts.
18:39
We have Washington, DC, we have Barcelona, we have Philadelphia, so everybody’s tuning in on this one from Dallas, everybody seems to be a group. This is really a global exchange fully. But aren’t you take it over from here?
18:56
Yes. Thank you very much both. Thanks, everybody. Thanks relocated with Research Alliance for inviting me to be here.
19:02
The pleasure to share the stage with Nick.
19:06
Leap, I worked for as a Social Data Intelligence Specialist, or when the Consultant.
19:13
social tech company be irrelevant. globally from Mexico, US. Pakistan in Australia.
19:22
So we are playing on this field, the artificial intelligence to look profit since awhile.
19:29
7 to eight years, first with the more easier things. But now with a bit more complex ones, you love to analyze the social data, all the data generated by our main audiences, donors, prospects.
19:44
So, thanks, mic two.
19:46
So, today, we all go to this, to do, we’re going to go into the agenda, because there’s a overwhelming thing.
19:55
Let’s move through formal to Joe, and then understand if you are ready for AI to start. And I will share some cases aligned with the strategic mapping that we explored before to understand if you are, if we are ready for AI or not.
20:11
So, let’s start with this.
20:15
The form of the, the form, which is a thing, is it vary, the pressure that has in the sector for me, is the fear of missing out to the pathology. That the peer is, the social networks. So we are under pressure to understand everything that is released. It’s impossible to follow LinkedIn. Nowadays, it’s lots of new AI GBT applications.
20:38
They’re putting work like, like mushrooms, but it’s not fair because if you look to the trends detect trends for 20 23, the things that we can use on fund raising.
20:51
It’s so complex, goes from full stack development of beauty, data analysis, metaverse, AI. So what I recommend is to move from the phone.
21:03
Do the, to the gentleman, which is the joy of missing out. We don’t need to know wherever every of these concepts and things protect ourselves.
21:14
Opinion. The best imports will before us, but in Brazil and one of the big topics within the sector is the mental health of the fundraisers.
21:23
And this pressure to be on top of everything, it doesn’t help at all.
21:29
So, let’s see what she should, we are ready for there.
21:35
And the thing, when we think about AI, the most important thing is being back, two, Can you can remove, please, Nick?
21:46
And, I like to do to look to these pictures. This comes from the tougher and denote the very well known … of interests. By the first wave of the future. Of shocking.
21:57
So, if you see of the human mentality evolution in the humanities since the prehistoric do, nowadays we have a huge periods of agriculture, basically using some tools to work in the field.
22:15
The big revolution, the first one of the industrial one, What was the bulk optimization, then candidate for a digital era with the telephone TV? All of these things that may cause the will to communicate between each other.
22:28
And that comes where we are now: the genetic nanotechnology robotics.
22:34
So I would ask, Can we ask you to sink a bit?
22:38
Can you, Becky?
22:41
Sorry to see where your organization is on the scene of the technology, because it’s here.
22:48
It is from the industrial era, basically the CRM, automation of seeing the tool to optimize that. Of course, it generates lots of data.
22:58
Then, ask yourself, which I really digital, and hearing what listening, what people are saying.
23:04
Really communicate that, are strong digital presence. Use the digital technologies, and gathering data for the directions of the social networks to understand better the behavior of your audiences.
23:17
And then, it’s there. Yeah, but this is the journey.
23:21
And we need to be ready today, take care of our own datasets, because it’s much more interesting and fruitful.
23:28
You should do the questions to our data to the LLVM. Rosette is important to understand our, they can use they had over our data.
23:38
Because our problems and probably the question, the ranch with what our core problems, they are in our state.
23:46
Nick, please.
23:49
Bringing this is the Chasms theory, which is the in tech startups, you look at this is the stage is the technology product, and we have different kinds of clients.
24:02
So are you early market?
24:06
She’s a tech enthusiasm, vision. There’s a bunch of innovators and early adopters, or this is not your organizational culture.
24:14
Probably, maybe you can, maybe you’re pragmatic, or even the conservative, nutcases freedom of truth to take the decision.
24:22
So it’s important to figure out your organizational culture. Where do you fit here?
24:30
And your information systems to understand, the key allows you to start doing AI in the more professional way, instead of just fishing, please.
24:44
So the first thing to understand is what problem I want to solve, because as Nick showed, there is AI for it.
24:52
I like to do this to use the balanced scorecard, the classical work.
24:56
Because, I feel very cool, comfortable with, we have a classical framework.
25:01
And then, we have some technology but this is the strategic map created by Clapton then the Lord, basically.
25:11
Nick, it organized a strategy for perspective initial one.
25:20
Processes, learning and growth intelligence.
25:23
From these four perspectives, we can build the strategic map, for instance, if we want to Yes. Good.
25:34
Oops, Nick.
25:38
If you want to do other stuff to improve your lifetime value, go, can we do it?
25:43
Reaching new audiences are retaining existence, but you need systems, the system generate data.
25:49
Then we have all the basic Organizational culture or culture into those individual soft skills.
25:55
So, when we do, we have a yeah, to help us. And, all of these things, this one just says, it’s a fake one, But you can use it in different in different contexts.
26:08
The next one, please.
26:16
Yeah. I mean, now, imagine if you want to improve their life than just putting the formula. You can use different ways to calculate the left until this one. Basically, what we want to do is to improve our donations.
26:28
We presented by ID and the Reduce our ….
26:32
So when we tried to do this deconstruct worries, country boot can impact on your donations. We can have different payment success rate or first relation possessing recurring donations.
26:49
We can create cross sell, or increase the donations, help, sell, dollar upgrade, whatever we do.
26:57
Then if you, from this your slides, you can start to understand where is your problem, at least it make you think about what do I link to? Where is my problem to address?
27:09
Me, please, in the same for the tuition. So we have been activations. And cancelations will be the way that the donors left. The inactivation is the business, the business rule. We decided to inactivate our doesn’t.
27:22
It doesn’t, doesn’t make well donation in the last six months or receive a call or e-mail asking us to conserve it.
27:31
Did you see the thracian can be in one of those breakthroughs? And then we’ll select if we understand which one is impacting our attrition, we can find that AI for good.
27:44
So and then even even when we go a bit deeper, the output is those zip codes.
27:51
And once the payments accessorizing activations, basically their transactional operations.
27:57
So the kind of technology that we use is different there.
28:01
The other one, to analyze behaviors, like, what made someone do the first donation, the recurring donations, can understand who wants to write.
28:13
So, just the understand our problem, and then you can go to the use case.
28:22
I’m bringing just a few ones.
28:26
We’ll start with the VCs, or someone from, our clients. are the ones from our colleagues. This one is the jet, our good friend from Argentina.
28:38
You’ve used AI to understand?
28:42
Yeah.
28:44
But the truth, to do a great donation, that, if they understand with the machine learning model, that the dollars would be more than one it, there’s much more purposely too.
28:56
Great.
28:57
But in the database, use it, there was a lot of move, no information about the number of kids.
29:06
So AI was used to build the new fields, guessing the name, the number of kids, at least one, and then they did the grid.
29:17
But it’s a way that the AI was used, two guests to predict the number of kids in each donor.
29:28
Next one, Nick.
29:31
The key thing is, do we use a …
29:39
supervisor AI models. The K means to understand that overlap on the blue cluster and the cluster is the number or the dollars that probably belongs to other clusters so we can upgrade.
29:57
Next one.
30:00
My good friend Christopher Holes allow it to the UNHCR, basically the needed tests on the net conversion rate to set the Facebook Ads playing with the with the dominant emotion of each hand.
30:16
Interesting, that the sentiment analysis in the emotion analysis is one of the main three legs of natural language processing, VCG, to choose the intersection between AI and linguistics.
30:29
We use that kind of technology to analyze entities, which are persons, organizations, governments, and topics within the topic, we analyze the narratives or counter narratives whenever you work with the.
30:46
here is a Conspiracy of Hope Horse, and that kind of information disorders. And then we have the sentiment analysis with chips in that. The still growing Indian comes to the demo page.
30:58
Analysis is still a film like that, needs a lot of improvement.
31:03
But you can see the relation to the net conversion rate with the demographics, Generation Y, X boomers and Civics.
31:12
And you can see, for instance, in the first column, the sadness works very, very bad, degenerating X But worked better with boomers.
31:23
We are analyzing the net conversion rate.
31:26
So it’s a way to use a bit of the sentiment analysis crossed with demographic data.
31:34
To identify the, the motion that performs better, concerning the net conversion rate.
31:42
And then we can develop our the next one, please.
31:48
Sorry, Phillipe, can I ask a question here?
31:51
Course, I mentioning here. Sadness was kind of the coal. Coal, French, the call to action.
31:58
Also be written in a sad way, or the story of the program that you, that to share. Can you tell a little bit about, how do you define the set, the spot, what, what was behind it?
32:11
It’s basically, the algebraic meets the spirit that are … as an explorer played before.
32:20
A read, they did all the processes bleeding, tokenize.
32:24
It didn’t read the head and compare with other ones, which is that, which are radically 65 from all the way.
32:34
Yeah, This one, the dominant emotion of this piece of text, or image, is possible to do with images.
32:45
It’s on the glass thickness.
32:49
That’s the way the talk.
32:51
Yeah, thanks, Tend to thank you for asking again. this is one of the kids, interstitial Galoshes.
32:57
In Brazil, it’s too, the main donor acquisition show was the face to face or district covers for the US market.
33:07
Then we started facing summer, can normans the tuition, or will become low rate layer layer, and then we decided to analyze the tuition is to predict the vision, right, for different cities, where we were running the campaign.
33:27
So it’s other way to make that decision.
33:31
Something that I forgot to say.
33:32
In the beginning, I worked, as a fundraising, directed, mobile, fundraising advisor, different Google, and do the thing that made me make the decision to start working we did, is look at the moral dilemma with the CPG, with the cost per donor.
33:49
Always lead for this problem, to remove the shame, the organizations are paying for the dollars.
33:55
So my drive was to use a Yeah, to reduce the cost per dollar and improve the life of my Crusade, that’s why I’m using this kind of stuff.
34:06
Nick, the next one, please.
34:09
Just wanted to bring this mythical, The Amazing Internal Team. What they did was to predict the two.
34:16
They use machine learning to identify the best day to charge the alerts and the best gateway payment gateway.
34:23
It’s just for each kind of like Visa, mastercard, amex.
34:28
And so the improvement was 11%.
34:32
Nick, the next one, we are almost finishing this bit, is a more sophisticated one, we did it, the …, we use the Not …, we use the …
34:44
20 12 to 1 first, with API when you’re using goats to without the chechen seasonality, but the technologies available available for a while for the commercial sector.
34:56
I believe that there’s the good for lots of time.
34:59
So basically, we use AI over graph technology to and we tend to do a job for the cop 27 when report and deployment changing.
35:11
And we found a cluster of the climate activists and the climate change deniers.
35:18
You will use the once we identify both clusters.
35:21
That’s why it’s important to workout with him with our own data.
35:24
Because this is our out if this is the, we collect data for social networks, mostly Twitter, YouTube, comments. And so. And we found some clusters. And then we compared those ones.
35:36
The can we use the GPU for two to build counter narratives to deployment deniers?
35:48
Comparing the two datasets. So basically what we do with our, we are very happy when they saw the Microsoft co-pilots model, which is the application of the AI engineers on the Microsoft office should be the Office 365.
36:05
And basically, what you do, this Microsoft model, the main data sources, they are there, the words, themes, PowerPoint, Excel, own data.
36:21
Then you can do magic. We can ask for a given document. We can ask the co-pilot to to build. The presentation. is 10 slides for a given elegance.
36:31
Though, or from a, or to produce an Excel spreadsheet. So that combination, that’s the technology that we use in house, not, the co-pilot because the same model we use social data.
36:42
It can be from social networks or whatsapp or telegram, depending on the data that we have available, or even service, large language models. And the craft technology, basically, the technology sensing that I learned once.
36:57
And almost 10 years ago, we were busy doing some startups in San Francisco, dress form of the traditional rows and columns, database, in nodes and edges.
37:09
And build a much more rich understanding of the data to the connections between the the notes.
37:16
Then, we can identify the underlies.
37:19
The Hedges depending were the size of that.
37:24
The next one, please, Nick.
37:26
Just to echo that for a second, the need for own data is absolutely vital. If you go into … off the shelf and say, write me a fundraising letter, it’s going to try to write you the most letter letter. That ever lettered.
37:42
But it doesn’t know that you have to put a PS out, and it doesn’t know any facts about your circumstance.
37:48
It doesn’t know what makes an effective letter, And so if you’re training out on, here are the things that I’ve sent and here are the results, then it can optimize for what works, not what it is.
38:01
And that’s a critical component to be using your own own data.
38:08
I do have a question from one of our our Westerners question. We are a progressive non-profit, committed to advancing the artist to careers of women, transgender, non binary and intersex artists. How can we reconcile the use of this technology which seems to be counter to our mission? They’re thinking thinking there, is that up, the concern is that generative AI will create art instead of artists.
38:38
And I would say that’s definitely a challenge.
38:40
I would look for some of the ethical AI systems that are existing, so for example, Adobe Firefly.
38:49
They’ve taken an interesting approach where they are using only images that they have and have licensed, so they’re not training on any sort of outside data.
39:02
So, they’re respecting artist rights and that sort of way, and so I would look for tools that are trained in that way. And I would look to train your own tools in that way.
39:11
The other part is that you can also be looking at how this can be, an advantage for you in the marketing of some of the artists that you are looking to elevate.
39:24
So, much of the barriers that we have, our time, and you may not wouldn’t be able to create all of the content that you want about a piece.
39:35
And, one of the things that some of these, even off the shelf shelf solutions are very good at, it’s summarization taking something longer, and bringing it down.
39:43
And so, you can take the statement that an artist has made about a work, and use that to summarize, for a LinkedIn post, or a Twitter post, or a Facebook post, those sorts of things.
39:54
So, they are still showcasing the art that humans do, in a human way, but taking some of the time out of the equation so that you can do more for them.
40:06
I would also say there’s a very interesting space in human AI generated art right now that, um, all one of the great things about this is that these tools can help people who have challenges in some areas.
40:25
My kids are both on the Autism spectrum and it has helped them with their fertilization.
40:31
And so it can be a way of helping open doors, or people in different ways. And so you may have find someone who embrace it and want to go that direction.
40:44
Given during your last example there, could this be something that is more, or can be used for service delivery as well, not just creating fundraising?
40:56
Abbott: Absolutely. And let me defer that question over to my colleague, because she has some good examples on program delivery.
41:05
Oh.
41:07
Think I can look for this question because it seems like it’s actually something that they’re working. one of the things that the question is, how do you leverage such a pity and keep your data within your environment rather than giving proper data to the larger model?
41:23
So, exists in the Summer Information Information security rules.
41:32
Then it should be comply with, for instance, the easel, that to me, is the, it’s the easiest thing for them, 27 zeros there one.
41:44
So it’s important to see if you’re with the providers that they are providing you the …
41:51
technology with your own data, to do a due diligence and to ship, which they are comply with the some references. This one.
42:00
The user Diesel 27001 is 1 of them there, but exist other kind of licenses licenses.
42:11
But, the data privacy standards, that it is, this is something to be aware, because with the motive start ups, the third thing we’re providing this, kind of, we’re offering this kind of service something that is really important.
42:29
Do you see the data privacy contracts, and the small letters?
42:37
And, of course, it’s even better if you can ask for a lawyer to do to check it in the C P inverse to comply.
42:49
We, can, we, can, we? Can. We move forward so I have just a couple of slides just to, to bring that. No, I’m bringing this Bernie Sanders, the first one.
42:57
The very ones, this tweet from, barely centers, and I think it’s pretty amazing, well, why do we want to do, to use AI.
43:06
This is amazing that in the fifties, the co CEOs were paid 20 times more than the average worker in the eighties, 59 times more.
43:16
in 2009 ended 80 times more, today.
43:22
They are paid almost 400 times more.
43:26
So, let’s see, the next slide.
43:29
I produced the the graph with these figures and it’s pretty similar to the next one that we saw.
43:42
In the beginning. Look at the curve is pretty similar.
43:45
So, can we, can you use the AI to, to raise more funds?
43:50
Can you we use the Day two, to produce more wealth?
43:56
give more, to get more donations? Because this is one of the commercial sectors.
44:02
So I love this tweet from Bernie Sanders that we can see OK.
44:06
Probably over the thames’s evolution from the industrial here, which is the or the fourth digit over to the fifties with the or even before the boom of the TV.
44:16
The group is pretty similar.
44:18
So we want to use a helpless sort of our programs to generate more impact.
44:25
And, two, make our co-workers feeling better and work in the bedroom. So, I’m throwing these, they’re bringing now some concept that we are implementing, key way to use AI.
44:37
The next one, please.
44:38
So, it just opened a new company here in India already in absurd. Alexa. Can say Alexa. Sorry about it. Will meet me. So. Our AI can get in the way of things.
45:05
Jessie, each can just unplug Alexa.
45:11
Sorry, Alex, who’s watching us sort of, so. So, we are implementing these, 4, 4, 3.
45:20
I won’t, the men in my company will work four days per week, per hour, per day, in three weeks, per month.
45:27
This is the game and keeping the same level of settlers.
45:31
And how can we use?
45:34
And I asked, Hey, I can make it happen.
45:37
Andrew.
45:41
AI, Explain the simple sketch, incomplete.
45:45
We can reduce the onboarding time of our clients by 20 to 30%, on the product garden and save 20 to 10 to 20% of administrative efforts.
45:57
Next one, Nick regarding the, then only on the nutritive tests and we have a lot of administrative tasks in different ways.
46:10
From the retention program to donation processing, the reporting, there’s lots of that. We can save 30 to 50% of our employees.
46:22
… communication system, OK, the kinda go to the last one.
46:27
So, just an example, why we are using AI, one of the things can be, simplify our processes and make it a lot more time to creative thinking, strategically thinking, to look to our data, to do this process, the deep analysis, who understands where really we are.
46:52
and the and just finishing duty is my last slide and then, thanks and we are hoping for or any questions on the remaining time.
47:01
My contacts are here to reach out.
47:06
We do have a couple of, we do have a couple of questions here.
47:10
Do any of the panelists know any technology to introduce, uh, oh, sorry, I mean, just runtime.
47:21
Do you know any technology to produce or insert our own content or data into generative AI?
47:28
You mentioned, show.
47:31
Chaired, the very, very good when the pet of the work was developed by the empty.
47:35
It’s, it’s compliant with the several Security Standards.
47:41
So, this is the, a good example.
47:48
Yes.
47:48
And the, the, the possibilities of those types of AIs are, in my mind, even greater than in Chachi PT or Dali, or any of the commercially available ones. What?
48:05
what is interesting is that Google and Microsoft, and folks like that, don’t have a moat around these … solutions. Very well.
48:14
And so there was A programmer who looked at the, the political biases of Chachi PT.
48:22
And here in the United States it would be considered left of center.
48:25
Libertarian said, what if I said it is the same system training data from prominent conservative thinkers.
48:34
And he was able to generate what he called right Wing GCT.
48:38
That is exactly opposite on the spectrum of GBT itself.
48:42
And he did it or under $200, using open-source tools and essentially training time from Amazon Web Services.
48:54
And so there’s a great deal of possibility in those types of open-source tools that.
49:04
four argues.
49:11
OK, we’ve got another question. Hello, we are an Alzheimer’s Research Foundation. We have a blog with very qualified content for families with Alzheimer’s Patients. How should we approach the fact that AI tools use blog content to generate responses, without any credit or reference to the source used?
49:38
This is a massive, massive challenge.
49:43
Um, back when the web was green, there was A discussion of whether newspapers would allow Google to spyder their sites and use their content and link to them.
50:00
And newspapers said, why, Why would we not do that? We’re getting the advertising revenue as it comes in.
50:07
And you can see what did decentralized model has done to those sources of information, as Google, in particular.
50:16
But all search engines have been bringing more and more material that are zero click queries, that you can go directly to that search engine page and answer without clicking through.
50:30
That is going to be even more of a challenge with AI, because it’s difficult to say where any AI idea comes from, it’s difficult to attribute, and it’s just like it is for us, humans. You, how do you know this piece of information?
50:48
Well, it, I learned it first and sixth grade, and then it was brought up on this college course. And then I had to use it two weeks ago. And so I remember it. But all of those different touch points, there’s no copyright for my sixth grade teacher for that idea.
51:04
And so it is a significant challenge and I would say that the big thing that you can do if you want to is to put the no robot’s text onto those pages so that they will not be spyder by search engines or AI. However, that decreases search ability of those engines. And so it’s a, it’s a conscious, strategic choice of you.
51:26
Do you want to maintain your Enclave? Or do you want to go completely the other direction and say, I want to get as much information out there as I possibly can. That’s from arch, and be the training data.
51:40
To say, this part of my mission is to educate, and whether it comes through our blog or through these other engine, some agnostic, too, that’s a tough choice.
51:56
Thanks, Nick. We have a couple of more questions coming in, so we just keep going, if that’s, if that’s OK. The next question is quite interesting. Do you have any best practices in responding to folks who want to use a notetaker in meetings instead of attending?
52:11
Am making all of a sudden meeting? Feel uncomfortable because it is being recorded without their consent?
52:17
Oh, wow, I would like to have a go, Felipe. Nick?
52:22
Yeah, I think it’s a It’s, the issue is the, It’s about the data overload, governance.
52:30
Yep.
52:31
It says that, Well, one of the things that we should consider to ever open data policies in the Greek Yeah.
52:40
The level of decision is the level in the data governance, and everybody should agree or not, but, probably, to seem to work within two, or to deal, or short-term, within the organization or the company.
52:58
For the Philosophical …, right, of it’s not right.
53:02
But seeing that, the decision needs to relate, it is related with the data culture in the data governance policies.
53:12
Yeah, and I was almost argue, this might be actually an interesting kind of cultural change, as well, because if you’re really looking at Europe, where TDP GDPR has been 14 years ago, and really change that behavior in a way, as well as asking permission, reporting data, continuously, it’s a very normal thing to do.
53:34
So, I was almost to ask a question back of house. Yeah, what does this say about your openness, flipping your team? How do you make agreements within your team? because I think a lot has to do darussalam.
53:45
Also, shutting ends, the common common understanding about this.
53:50
The only I’d make off of that is, if you can get the same content from your meeting by reading the transcript, then it doesn’t need to be a meeting. It can be an E.
54:02
And that brings us to, how many meetings do you need? And that’s almost goes back to Phillip. Is how can we get to a 4, 4 3 system. So that’s one thing, definitely, that we know from a social impact sector. And enough analysis has been done by our deaths.
54:19
That’s our meetings are not always the best and the most functional ones.
54:23
So yeah, do we really need a meeting for it? Are there are many blocks went on around depths? Let me move to the next question.
54:30
How soon do you feel that Google’s algorithms will be able to identify and polish impregnates companies and organizations for using AI to choose copy on the websites?
54:44
The tricky one, because you can we can ensure it’s workflow from different perspectives. I think Google has already the technology available, for sure. And I’m pretty sure there are doing it.
54:56
Even though they know they know it’s good, easier to do, can use AI to the cafe, the signal will be real more related to the object or more freedom of expression.
55:12
Can Google punish anyone two to publish or not?
55:19
That’s the world to think of archaic.
55:22
The thing that’s really scary though is, I don’t know if anybody saw the 60 minutes apiece over the weekend 60 minutes as a news program here in the US. But then people from Google and other companies and frankly what they what their what they were talking about was very frightening.
55:38
What they can do right now, but how they’re going to implement it slowly as as humans get more used to each part of it, though we roll out some more of it, and I was pretty hair raising.
55:55
But then I would also say that Google’s algorithms, to some extent, already can detect AI generated content.
56:03
And there are some penalties in place. The fully generated content is fully against their guidelines.
56:10
But they are EITI sorry, ET System, for how content is ranked.
56:20
We’ll also take down those things that don’t have expertise, and authority and trust behind them.
56:28
So Google, in theory, is already doing that in practice.
56:32
The analogy that I always use for this is that AI should be, the partnership between human and AI, should be like the partnership between the human and device.
56:42
It’s a tool, it’s a tool for helping you go faster, and it’s better when you work together.
56:48
AI generated content, no, AI helped content, yes, I just finished the second draft of my book, and definitely having AI go through it for grammatical changes in addition to a human proofreader.
57:02
I think, to, The human element is so important, still. Eventually, it might not be, but, still has spotted columnists that we’ve had.
57:13
Spotted some phrasing that was not quite what that calmness had given us before in terms of every day.
57:20
So I asked them individually, did you use AI for this?
57:24
And they said, oh yeah, I was a little busy so I, so we can now policy at the non-profit times that nothing and our columns can be generated with AI and if we catch and they’re no longer columnists.
57:39
I think that’s also reflects back to watch Filippo also mentioning about, you know, if you’re a photographer reflecting that data analysis, what is the quality of your data?
57:48
Because that’s really where it all starts. To be very honest.
57:53
And we had a conversation about, really, when we were preparing for this session, we have done research last year to explore across NGOs from the cloud.
58:02
Quite, a lot of them were based in Europe and in the US About hood, it’s well, you know, really, really adapting to watch Colpitts and new techniques to use and new tools, and new ways of reaching out to audiences and who didn’t. And I founded really shocking when we found out it’s only 25% of the organizations. And people interviewed said, we really know who our supporters are. Because we have to live technical tools in place. And therefore, we could adapt liquid depth plus by testing, understand the results, and immediately saying, hey, stop or continue, or adapt accordingly again.
58:39
But definitely, it’s a 75% of the organizations in our …, of which half of them said, We do not even have to start with our technology.
58:49
So if you translate that now to AI and using your data in the analysis of AI, it’s almost coming back to this graph, where I probably that’s first step needs to be in place before utilizing all these other opportunities that are there.
59:06
Really wanted to highlight that, because we often jump immediately to that.
59:11
latest latest new technology, but facing in many ways, also coming back to the basics that we understand our audience is do We know what works and then we escalate no systems and tools in place to start knowing this website. I’m sorry, is there a place on your website where people can go to go and see that research?
59:33
Putin, I put a link, and it’s all we can maybe send it a message to afterwards, as well.
59:41
Nice. Somebody asking for specific AI tool. Of course, we have attempted to implement workflows for repetitive processes, but you haven’t found the right tool yet. Is there one that you would recommend that will provide some intelligence workflow?
1:00:00
Any recommendations, Sneakily?
1:00:03
Oh, Thank you.
1:00:06
We didn’t reach the point where we can exist, the solution already.
1:00:14
Ready to work out the full walk workflow, What we are testing is using the simple workflows, tools, like made dot com, or make dot com, … dot com. Even the table, and then you can assemble the D three.
1:00:35
AI calls for different processes.
1:00:38
I don’t know, one, that is already no already ready.
1:00:49
Work with the workflow cause.
1:00:56
At, and I think it would depend largely on the workflow that you’re looking to automate.
1:01:05
Um, the one that I have heard of is, there’s one called the Forge that is four producing app that create workflows and processes. That’s supposed to be agnostic.
1:01:20
But, if you go to, There’s an AI for that, There are very specific ones to say. Let’s automate this sales workflow and it has some of the materials in there and those sorts of things.
1:01:32
Maybe it does, I’ll show you … Paul.
1:01:34
Yeah, we got 1 last 1. Go ahead and take it.
1:01:37
Yeah, I’ve talked about interesting one, because it says. The last question is any good training materials, using AI in MTL fundraising. I’m going to do a tiny.
1:01:48
That’s a tiny link here. If a masterclass Stefano organizing in August 22nd, and 24th of August, the organizing to … to be same. But they are different time zones, which is literally called AI fundamentals because we have very high demands of understanding more about it. So of course, that’s one of the ways to do that, Maybe philippa make any recommendations from your end, where people could go to, I should have a look at, to understand more about AI in profits contracting.
1:02:22
I think is the world of, I don’t know, too much content already pictures for the sector, by, But the thing is, they say, It’s about your problem.
1:02:34
We’re just in the moment that we can take an idea from the commercial sector.
1:02:39
There’s 10 inspiring cases, just go onto YouTube, YouTube firm, and tick tock.
1:02:48
Go to the to the talk and make your questions.
1:02:52
There are exempt below the Coca Cola is using using the AI to improve sales.
1:02:59
No because we are very the baby steps, small them to get inspired and replicate context once we load the problem we want to address.
1:03:13
And that actually is the topic or my upcoming book hope to get it launched in the next few weeks.
1:03:20
But in the meantime, it’s not non-profit specific, but Microsoft and LinkedIn have a career. It’s essential in generative AI through LinkedIn learning. That’s free.
1:03:31
And then Google has a number of free courses available through through Google Cloud.
1:03:37
If you Google Generative AI training, you’ll see levels of introductory, intermediate advanced.
1:03:46
And some of them are on TensorFlow, which is the open-source product. That’s really helpful. A lot of these open source, as well.
1:03:56
We’ve covered a lot of ground today.
1:03:58
I’d like to thank our speakers, and you want to tell them a little bit about the Resource Alliance, where they can go get some more information.
1:04:08
Well, for anyone wants to see more about our events, we organized a lot of free, free, monthly webinars around different topics, not only about AI, but to me, ask our community about what are the challenges for you to, you know, to reach and achieve impact. And those topics are investing monthly webinars. But also, what I was referring to when we lose some more in-depth analysis and training series, and is AI, one is definitely coming up in August. On these topics, that’s almost changes on a day-to-day basis. Let’s see where we are in August. You will find the links that will be shared in the chat.
1:04:46
Some, some overviews about what is one thing. But of course, very happy to update you as much as well. And so, yeah, it’s definitely a big topic as an international filtration congratulating over where people are continuously remembering, reminding me, like they cannot give any kind of hand out of the session. Yes, because October is so far away, AI wise that we don’t know yet whether they won’t be deaf. So, yeah, we’re very excited to have thoughts on the agenda.
1:05:14
I’m, of course, more open and free information sharing.
1:05:21
Thanks for being there with us.
1:05:23
Give us, or give it. Give everyone your website.
1:05:27
Yes, WWW dot resource, this alliance dot org. And we put it in the chat, as well.
1:05:37
And please, when you get a chance, go to the non-profit Times website and subscribe to the non-profit Times, Read the largest business publication, and business, and fundraising publication in the sector. More than 200,000 people get something from us. Either print and print to digital every month. So, please, go to WWW the non-profit times dot com and take a look. Again, thanks to our speakers. Thanks to everybody who registered. And we’ll see what the next research alliance, a non-profit times webinar. Thanks for coming.
1:06:09
Thanks.
